Achieving global carbon neutrality targets requires innovative approaches to improve the energy performance and thermal comfort of buildings. This study investigates the application of Physics-Informed Neural Networks (PINNs) for predicting indoor thermal comfort conditions in residential buildings, combining physical modelling with Machine Learning (ML) strategies. A Multi-Storey Residential Building (MSRB) constructed in the 1980s in Bari, Southern Italy, is selected as a case study. A dynamic energy simulation is performed using DesignBuilder software, calibrated according to ASHRAE Guideline 14, using real weather data. The generated data set, including cooling loads, solar gains, occupancy, ventilation and outdoor air temperatures (OATs), is used as input to the PINN model, which directly incorporates the Energy Conservation Law during training. Hyperparameter Optimisation (HO) is performed using Optuna, and a PINN ensemble strategy further improves prediction accuracy. The final model achieves a Root Mean Square Error (RMSE) of 0.035 °C for the prediction of the operative temperature and allows the estimation of the Predicted Mean Vote (PMV) with a comfort compliance of 2.60%. Compared to the traditional PMV calculation based on measured temperatures only, the PINN-based approach shows a significant improvement of approximately + 0.66% in comfort hours. These results confirm the ability of PINNs to provide physically consistent, data-driven predictions of both thermal behaviour and occupant comfort, providing a robust framework to support energy-efficient and occupant-centred design strategies in the residential building sector.